Electronic and optical properties of NbO2Andrew O’Hara,1 Timothy N. Nunley,2 Agham B. Posadas,1 Stefan Zollner,2

and Alexander A. Demkov1,a)1Department of Physics, The University of Texas at Austin, Austin, Texas 78712, USA2Department of Physics, New Mexico State University, Las Cruces, New Mexico 88003, USA

(Received 10 October 2014; accepted 17 November 2014; published online 3 December 2014)

In the present study, we combine theoretical and experimental approaches in order to gain insight

into the electronic properties of both the high-temperature, rutile (metallic) and low-temperature,

body-centered tetragonal (insulating) phase of niobium dioxide (NbO2) as well as the optical prop-

erties of the low-temperature phase. Theoretical calculations performed at the level of the local

density approximation, Hubbard U correction, and hybrid functional are complemented with the

spectroscopic ellipsometry (SE) of epitaxial films grown by molecular beam epitaxy. For the rutile

phase, the local density approximation (LDA) gives the best description and predicts Fermi surface

nesting consistent with wave vectors that lead to niobium-niobium dimerization during the phase

transition. For the insulating phase, LDA provides a good quantitative description of the lattice, but

only a qualitative description for the band gap. Including a Hubbard U correction opens the band

gap at the expense of correctly describing the valence band and lattice of both phases. The hybrid

functional slightly overestimates the band gap. Ellipsometric measurement is consistent with insu-

lating behavior with a 1.0 eV band gap. Comparison with the theoretical dielectric functions,

obtained utilizing a scissors operator to adjust the LDA band gap to reproduce the ellipsometry

data, allows for identification of the SE peak features. VC 2014 AIP Publishing LLC.[http://dx.doi.org/10.1063/1.4903067]

I. INTRODUCTION

Transition metal oxides exhibit a wide range of interest-

ing phenomena such as ferroic order, superconductivity,

interfacial two-dimensional electron gases, and metal-to-in-

sulator transitions. In particular, the metal-to-insulator transi-

tion has been observed in many different oxides;1–7

however, the transition temperature is often significantly

below room temperature making their use in technological

applications challenging. Recently, there has been significant

interest in the near room temperature Mott-Peierls transition

of vanadium dioxide (VO2).8–12 The closely related niobium

dioxide (NbO2) exhibits a similar transition but at a much

higher temperature,13,14 which may be of interest in both

high temperature applications or applications where the

effects of temperature need to be isolated from that of elec-

tric fields. In fact, there has been some progress in the appli-

cation of NbO2 to electrical switching15–17 and memory

devices.18,19

In VO2, the metal-to-insulator transition occurs at

340 K,20 whereas for NbO2, the transition occurs near

1080 K (Refs. 13, 21, and 22) accompanied by a structural

transition from the undistorted rutile structure (P42/mnm

with two formula units per cell) at high temperatures shown

in Fig. 1(a) to the body-centered tetragonal, distorted rutile

structure (I41/a with 32 formula units in the conventional

cell)23–25 at low temperatures shown in Fig. 1(b). The

Brillouin zone symmetry of the phases is shown in Figs. 1(c)

and 1(d), respectively. In the ionic limit, the niobium atoms

are in the Nb4þ oxidation state with a valence configuration

of 4d1. Given that pairs of niobium atoms dimerize along the

c-axis during the phase transition, the transition is believed

to be of the Peierls type (such that a quasi-one-dimensional

chain of dimers forms along the c-axis) with an instability

related to wave vector qp ¼ ð1=4; 1=4; 1=2Þ (occurringbetween the A and Z points in the Brillouin zone shown in

Fig. 1(c)); however, the exact nature has been debated

throughout the literature.22,26–31 Understanding the band gap

of the insulating phase of NbO2 is of both practical and fun-

damental importance for device applications. Interestingly,

niobium dioxide’s band gap is not precisely known and

remains an open question. Within the experimental literature

on the electronic structure,17,32–36 the band gap of the low-

temperature phase of NbO2 ranges from a room temperature

estimate of the optical gap of 0.5 eV (Ref. 32) to 0.88 eV

from film absorptance edge measurements17 and admittance

spectroscopy measurements33 to 1.16 eV obtained by fitting

to conductivity measurements.34 Recent measurements uti-

lizing a combination of x-ray photoemission spectroscopy

(XPS), ultraviolet photoemission spectroscopy (UPS), and

inverse photoemission spectroscopy (IPS) on high-quality

epitaxial films indicate the band gap to be at least 1.0 eV.35

In this paper, we use density functional theory to explore

the electronic and optical properties of the high- and low-

temperature phases of NbO2. We employ both standard DFT

and several extensions, in order to better understand the na-

ture of the band gap and possible role of electron correlations

in the insulating phase. Through calculation of the carrier

concentration, we give an estimate for the change in carrier

concentration across the transition. Finally, the real and

imaginary dielectric functions are calculated from first prin-

ciples and measured using spectroscopic ellipsometry ina)demkov@physics.utexas.edu

0021-8979/2014/116(21)/213705/12/$30.00 VC 2014 AIP Publishing LLC116, 213705-1

JOURNAL OF APPLIED PHYSICS 116, 213705 (2014)

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.219.247.33 On: Tue, 20 Jan 2015 02:57:55

http://dx.doi.org/10.1063/1.4903067http://dx.doi.org/10.1063/1.4903067mailto:demkov@physics.utexas.eduhttp://crossmark.crossref.org/dialog/?doi=10.1063/1.4903067&domain=pdf&date_stamp=2014-12-03

order to understand the optical absorption and give an inde-

pendent measurement of the band gap.

II. COMPUTATIONAL METHOD

We carry out density functional calculations using the

Vienna ab initio simulation package (VASP) code.37 For theexchange-correlation functional, the local density approxi-

mation (LDA) parameterization by Perdew and Zunger38 is

used and projector augmented wave pseudopotentials are

used for niobium and oxygen electrons.39,40 The valence

configurations for each atomic species used are the

4p65s14d4 orbital configuration for Nb and 2s22p4 orbital

configuration for O. A plane wave cutoff energy of 750 eV

was used for both phases, while the Brillouin zone was

sampled using C centered Monkhorst-Pack grids41 of 8�8� 12 for the rutile cell and 8� 8� 8 for the primitivebody-center tetragonal cell. These combinations of cutoff

energy and k-grid provided convergence of 1 meV per NbO2unit for the total energy in both phases. During relaxation of

metallic rutile NbO2, the first-order Methfessel-Paxton42

scheme with a sigma value of 0.17 was used for partial occu-

pancies, while the tetrahedron method with Bl€ochl correc-tions43 was used for self-consistent total energy calculations.

III. LATTICE AND ELECTRONIC STRUCTURE

A. Rutile phase within local density approximation

The optimized rutile structure for NbO2 within LDA

has lattice constants aR ¼ 4:93Å and cR ¼ 2:90Å versusexperimental values of aR ¼ 4:8463Å and cR ¼ 3:0315Å.25Furthermore, the optimized coordinate for the oxygen atom

at the 4f Wyckoff position (direct coordinate ðu; u; 0Þ) isfound to have u ¼ 0:289 versus an experimental value ofu ¼ 0:2924. The density of states plotted in Fig. 2(a) showsthat the Fermi energy lies in the niobium d-state-derived

band just above the charge transfer p-d gap of the oxygen p-

states and niobium d-states. Our band structure is consistent

with previous results.44–46

In order to understand the orbital origin of the niobium

4d states in NbO2, and in particular the t2g and eg crystal field

splitting, we compute the orbital projected density of states

for one of the niobium atoms with its octahedron of oxygen

atoms oriented along the x, y, and z axes (this is approximate

due to the Jahn-Teller distortion of the octahedron) in Fig.

2(b). Within the rutile cell, there are two different niobium-

oxygen bond lengths (2.02 Å along the z-axis and 2.064 in

the xy-plane for this octahedron). Furthermore, the bond

angles in the xy-plane are slightly distorted from 90� to

FIG. 1. Crystal structure of (a) high-temperature rutile (space group P42/mnm) and (b) low-temperature body-centered tetragonal (space group I41/a) phases of

NbO2 in the primitive cell. (c) The Brillouin zone for the rutile (simple tetragonal) phase where the suspected soft phonon mode wave vector qp lies betweenthe A and Z point. (d) The Brillouin zone for the low-temperature distorted rutile (body-centered tetragonal, c < a) phase.

213705-2 O’Hara et al. J. Appl. Phys. 116, 213705 (2014)

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.219.247.33 On: Tue, 20 Jan 2015 02:57:55

90.89� and 89.11�. As expected from the ligand theory ofcrystal field splitting, the t2g bands are located lower in

energy than the eg bands with a splitting of approximately

4.0 eV. Since each niobium is formally in the 4þ state, eachniobium atom has d-orbital filling of 4d1, which means the

degenerate t2g orbitals contain a single electron. Therefore,

the Jahn-Teller distortion of the octahedra causes a splitting

of the t2g level into a partially occupied dxy orbital and par-

tially unoccupied dyz and dxz orbitals in order to break the

degeneracy. Furthermore, the occupied dxy orbital for each

niobium atom is oriented in such a way that two of the lobes

point along the c axis (dimerization axis) towards the nearest

neighbor niobium atoms. In Fig. 2(d), the partial charge den-

sity for the niobium d-states below the Fermi energy is plot-

ted in the [110] direction. In the figure, a cross-section of the

two distinct dxy orbitals on nearest neighbor niobium atoms

is clearly visible, indicating that there is no niobium-niobium

bond in the rutile phase.

In Fig. 3(a), we show the calculated Fermi surface for

the full Brillouin zone, rather than just the irreducible wedge

as in previous work.44 Of particular interest are the sheet-

like segments situated just above and below the C-M-Xplane. The nesting of flat segments of the Fermi surface is

characteristic of one-dimensional conductors with Peierls

transitions.47 Despite being a three-dimensional material, re-

sistivity measurements have shown that metallic conductiv-

ity in the undistorted rutile phase is primarily along the (001)

direction, while the conductivity in the plane perpendicular

to the axis of dimerization remains semiconductor-like14

indicating that NbO2 is a quasi-one-dimensional conductor.

In Fig. 3(b), we show a diagonal cut through the Fermi sur-

face in order to investigate the potential for Fermi surface

nesting in more detail. From this plot, we see that there are

four pseudo-flat planes as part of the Fermi surface (with two

in the upper half of the Brillouin zone and two in the lower

half of the Brillouin zone). These four planes can be grouped

as an inner pair that exhibits a clear sinusoidal oscillation

and an outer pair that appears flatter on the cross section

with a slight bulge. For the inner pair, we observe that one

plane of the pair can be mapped to the other plane in the pair

by a vector of � 14; 1

4; 3

10

� �. Similarly, for the outer pair of

planes with the flatter behavior, an imperfect mapping of

one plane to the other can be achieved by vectors in the

range of � 0; 0; 25

� �to � 0; 0; 4:5

10

� �. While these vectors are

not equivalent to high symmetry vectors q*

P ¼ 14 ; 14 ; 12� �

or

q*

Z ¼ 0; 0; 12� �

, this is consistent with the notion of an incom-

mensurate or imperfect nesting of the Fermi surface.

FIG. 2. Plots for rutile NbO2 of (a) the total and orbital-resolved density of states, (b) the niobium d-state projected density of states, (c) band structure, and (d)

partial charge density for the d-states below the Fermi energy.

213705-3 O’Hara et al. J. Appl. Phys. 116, 213705 (2014)

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.219.247.33 On: Tue, 20 Jan 2015 02:57:55

Furthermore, neutron powder diffraction studies of the rutile

phase25 indicate large Debye-Waller factors. In particular,

the rms displacement for niobium atoms along the c-axis is

0.18 Å indicating significant fluctuation of the lattice vectors

and hence reciprocal space, which may improve the ability

to nest. It should also be pointed out that if in reality the flat

planes in the Fermi surface were perfectly flat, then any vec-

tor of the form kx; ky;12

� �would become a nesting vector.

This may help explain the presence of multiple soft modes in

the phonon dispersion calculated in Ref. 31, since the Peierls

picture of a phase transition links the lattice modulation vec-

tor (i.e., soft phonon mode in a second-order phase transi-

tion) to the nesting vectors of the Fermi surface.

B. Low temperature phase within local densityapproximation

We compare calculated structural parameters against

powder neutron data25 in Table I. Compared to experiment,

the lattice constants of a ¼ 13:64Å and c ¼ 6:01Å representa respective �0.45% and þ0.45% deviation, which is wellwithin typical deviations for such calculations. Furthermore,

the calculated niobium-niobium dimer length is 2.70 Å, com-

pared with 2.71 Å experimentally. As can be seen from the

table of positions, the primary discrepancies come from the

specific internal coordinates, in particular, those of the oxy-

gen atoms.

The density of states and band structure for the low tem-

perature phase are shown in Fig. 4(a) and the LDA band

widths and gaps are summarized in Table II. In terms of the

valence O 2p and Nb dxy state, we find reasonable agreement

with the previous calculation46 with a narrower p-d gap and

occupied d-band. For comparison, room temperature XPS

and UPS data35,48 show that the valence band width is

approximately 9.0 eV with the O 2p width approximately

6.0 eV and the Nb dxy state approximately 1.0 eV wide.

Compared to our values in Table II, this implies that the pri-

mary source of underestimation of the valence band is that

our p-d gap is too small. Our calculation shows that for the

LDA optimized structure a band gap of 0.35 eV opens due to

the lattice distortions of niobium-niobium dimerization and

tilting that occurs in going from the high- to low-symmetry

phases. Band structure calculations, shown in Fig. 4(c), show

that the gap is indirect, with the valence band maximum

(VBM) at high symmetry point N and the conduction band

minimum (CBM) at C. This value is higher than that foundpreviously in DFT with the augmented spherical wave

(ASW) (0.1 eV) and linearized augmented planewave

(LAPW) (0.15 eV) methods46,49 but lower than Nb2O10 clus-

ter calculations (0.68 eV).45 Our LDA band gap is, however,

still smaller than the range of reported experimental esti-

mates mentioned previously.17,32–34 The lowest direct band

gap (which is the quantity measured by ellipsometry) is

0.42 eV and occurs near the C point (more specifically at~k ¼ ð�0:1;�0:1; 0:1Þ and symmetry equivalent points).

Fig. 4(d) shows the charge density for the niobium

d-states below the Fermi energy in a plane containing the

dimerized atoms with their nearest oxygen atoms. This plot

shows the clear formation of a strong bond between the nio-

bium atoms through the dxy orbitals. This suggests that the

gap formation occurs with dimerization and not a further

Jahn-Teller distortion as initially discussed by Goodenough

for similar edge-sharing octahedral oxides.50

FIG. 3. Plot of (a) the full three-dimensional Fermi surface for rutile NbO2and (b) the diagonal cross-section of the Fermi surface with potential nesting

vectors.

TABLE I. Comparison of experimental and LDA calculated lattice parame-

ters for the body center tetragonal phase of NbO2 (space group I41/a, all ions

occupy Wyckoff position 16(f)). Experimental data are from Ref. 25.

Parameter Experimental Theoretical

Lattice vector a 13.696 Å 13.640 Å

Lattice vector c 5.981 Å 6.012 Å

Nb(1) coordinate (0.116, 0.123, 0.488) (0.112, 0.122, 0.475)

Nb(2) coordinate (0.133, 0.124, 0.031) (0.132, 0.126, 0.027)

O(1) coordinate (0.987, 0.133, �0.005) (0.986, 0.128, �0.021)O(2) coordinate (0.976, 0.126, 0.485) (0.970, 0.122, 0.509)

O(3) coordinate (0.274, 0.119, 0.987) (0.274, 0.125, 0.000)

O(4) coordinate (0.265, 0.126, 0.509) (0.262, 0.124, 0.502)

213705-4 O’Hara et al. J. Appl. Phys. 116, 213705 (2014)

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.219.247.33 On: Tue, 20 Jan 2015 02:57:55

C. Effect of Hubbard U within LDA 1 U

In systems containing d- and f-orbitals, both LDA and

GGA (Generalized Gradient Approximation) often underes-

timate the correlation of electrons in partially filled orbitals.

In order to account for this, a Hubbard-type correction U can

be employed for these orbitals. While LDA does capture the

correct physics to give a gap, from Fig. 4 we can see that the

bonding state is fairly narrow (Fig. 4(a)), relatively flat (Fig.

4(c)), and spatially localized (Fig. 4(d)). This implies that an

orbital dependent term may be necessary to increase the

band gap. Specifically, we make use of the rotationally invar-

iant method of Dudarev et al.,51 which uses an effective

Ueff ¼ U � J, combining both the Hubbard and exchangeterms.

To test the effects of using such a Ueff on the rutile phase

of NbO2, we perform calculations employing values from

0 eV to 8 eV. For no value of Ueff did an unphysical band

gap open in the rutile phase as occurs in rutile VO2.52

However for Ueff> 6 eV, spin polarized calculations showthat there is a ferromagnetic solution contrary to experimen-

tal observation. Furthermore, both the c=a ratio and cell vol-ume are closest to experiment when no Ueff was used in the

rutile phase as shown in Fig. 5(a). This implies that rutile

NbO2 is most likely an uncorrelated metal.

For the low temperature phase, we tested Ueff values

from 0.0 eV to 5.0 eV (higher values of Ueff led to conver-

gence issues) and the results are plotted in Fig. 5. In regards

to the lattice, there is a slight improvement for the niobium-

niobium bond as Ueff is increased, but as shown in Fig. 5(b),

the cell volume and c/a ratio are better in the vicinity of Uefffrom 0.0 eV to 1.0 eV and 1.0 eV to 2.0 eV, respectively. For

increasing Ueff, the indirect band gap plotted in Fig. 5(c)

increases from the pure LDA value of 0.35 eV to a value of

1.15 eV at Ueff¼ 5. From these results, it appears that minorcorrelation effects as included via the Hubbard U can

FIG. 4. Plots for the distorted body-centered tetragonal NbO2 phase of (a) the total and orbital-resolved density of states, (b) the niobium d-state projected den-

sity of states, (c) band structure with the valence band maximum (VBM) at the N point and the conduction band minimum (CBM) at C, and (d) partial chargedensity for the d-states below the Fermi energy.

TABLE II. Summary of the body-centered tetragonal phase band widths and

gaps (all in eV).

O2p width p-d gap dxy width Egap t2g width eg width

Ref. 10a 5.67 2.33 0.93 0.10 3.53 5.17

LDAb 5.55 1.58 0.70 0.35 3.41 4.87

HSE06b 5.72 1.79 0.71 1.48 3.28 5.32

aEstimated from density of states.bThis work.

213705-5 O’Hara et al. J. Appl. Phys. 116, 213705 (2014)

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.219.247.33 On: Tue, 20 Jan 2015 02:57:55

improve the description of the insulating phase, but that

other factors may be responsible for the underestimation of

the band gap within LDA. The application of Ueff up to

3.0 eV causes a slight increase in the relative value of the

lowest direct band gap to 0.10 eV; while for higher values of

Ueff, this relative value decreases to 0.06 eV. Throughout the

range of Ueff, the k-point for the lowest direct gap shifts

away from the C point towards k ¼ 12; 1

2; 1

2

� �. This implies

that the Ueff causes a slight change in the curvature of the top

of the valence band and bottom of the conduction band.

Multiple XPS measurements of the valence band of NbO2have shown that both LDA and hybrid functional calculations

reproduce the measurements relatively well.35,46,48 However,

when using LDA þ U, the observed increase in the indirectband gap of NbO2 occurs at the expense of lowering the width

of the oxygen 2p–niobium 4dxy gap as shown in Fig. 5(c). The

reason for this is that the inclusion of the Hubbard U will push

apart the occupied and unoccupied bands of the material, which

in this case are both comprised of d-orbitals. While the failure

of LDA þ U in this regard does not necessarily indicate thatNbO2 should be considered uncorrelated in the insulating phase,

it does expose a potential issue in applying this technique for

introducing correlations into density functional theory.

D. Hybrid functional calculation of low temperaturephase

Due to the lingering ambiguity in the literature over the

band gap of NbO2, it is useful to perform a hybrid functional

calculation of the electronic structure for the low temperature

phase. Hybrid functional methods use a mix of LDA or GGA

with Hartree-Fock since the former tends to underestimate

the band gap, while the latter tends to overestimate it. There

are several different hybrid functionals available, with the

Heyd-Scuseria-Ernzerhof (HSE) method53,54 being fairly

popular due to its ability to often improve the calculated

band structure.55 The HSE functional separates the exchange

part of the exchange-correlation function into a long-range

part from the Perdew-Burke-Ernzerhof (PBE) functional56 (a

type of GGA) and uses a mix of the Hartree-Fock exchange

and PBE exchange for the short range portion. The full

exchange-correlation functional is thus given by

EHSExc ¼ aEHF;SRx ðlÞþð1�aÞEPBE;SRx ðlÞþEPBE;LRx ðlÞþEPBEc ;(1)

where a controls the exchange mixing and l controls the cut-off for short range interactions. In HSE, the mixing parameter

a ¼ 14

was found through optimization of a wide variety of

systems. Specifically, we employ the HSE06 hybrid which

sets l ¼ 0:2Å�1. In practice, a standard DFT run with PBEbased pseudopotentials is done to produce a converged charge

density and wave function and then the hybrid calculation is

done with these as a starting point. While relaxation can be

done within HSE, for larger systems like the low-temperature

NbO2 body-centered tetragonal cell, the structure is optimized

within regular PBE due to computational constraints.

The previously used energy cutoff of 750 eV and k-point

grid of 8� 8� 8 for LDA were found to give similar conver-gence for the PBE pseudopotentials. Using the PBE optimized

lattice, we calculated the HSE06 indirect band gap to be

1.48 eV, which is higher than any of the previously reported

values in the literature and likely represents an overcorrection

of the gap. Most importantly, the band widths themselves,

shown in Table II, are quite similar to the LDA results above.

E. Carrier concentration and the phase transition

One of the important implications of metal-insulator

phase transitions for electronic applications is the change in

FIG. 5. Summary plots for the effects of the Hubbard U on (a) the high-

temperature rutile lattice, (b) the low-temperature body-centered tetragonal

lattice, and (c) the band gap (diamonds) and oxygen 2p–niobium 4dxy gap

(squares) in the low-temperature phase.

213705-6 O’Hara et al. J. Appl. Phys. 116, 213705 (2014)

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.219.247.33 On: Tue, 20 Jan 2015 02:57:55

conductivity across the phase transition. Conductivity can be

defined in the simplest sense as the product of the electron

charge e, the carrier concentration ne, and carrier mobilityle. In practice, the carrier concentration is easier to calculatethan the mobility, which requires a detailed knowledge

of scattering mechanisms, and so we focus on the carrier

concentration change across the transition. This focus is sup-

ported by the experimental observation34 that comparatively

the mobility changes less drastically than the carrier concen-

tration across the transition.

In the metallic rutile phase, we calculate the carrier con-

centration at a given temperature via

ne ¼1

V

ðf Eð Þg Eð ÞdE; (2)

where f ðEÞ is the Fermi function, gðEÞ is the density ofstates, and the integration runs from EF � kBT to EF þ kBTsince we expect only electrons in this energy range to be

thermally active. Using our LDA density of states, we obtain

an estimate of ne ¼ 4:57� 1021cm�3 at the transition tem-perature of 1080 K, which is slightly smaller than the experi-

mentally reported value at this temperature.34

In intrinsic semiconductors, the numbers of electrons

and holes at a given temperature are equal. Therefore, in the

insulating BCT phase, we can compute the number of elec-

trons using the same equation as above where integration

runs from the bottom of the conduction band to a suitable

cutoff (in reality, the upper bound is þ1; however, theFermi function minimizes contributions far above the Fermi

energy). The number of holes (nh) is calculated by replacingf ðEÞ with 1� f ðEÞ and integrating from a suitable lowercutoff (in reality, the lower bound is �1) to the top of thevalence band. We found that 65 eV from the conductionand valence bands gave suitable convergence for the integra-

tion range. Furthermore, at finite temperature, the Fermi

energy is no longer precisely at the mid gap level, so we

adjusted it self-consistently until we reach equality of neand nh. Computing the carrier concentration using ourLDA density of states at 1080 K, we obtain an estimate of

ne ¼ 3:71� 1020cm�3, which suggests an order of magni-tude change in carrier concentration due to the metal-

insulator transition. We also computed with our HSE06

density of states and found ne ¼ 1:02� 1018cm�3, whichwould suggest a change in magnitude of 3.5 orders of magni-

tude. While these numbers vary greatly, we anticipate that in

reality the actual degree of change would lay between these

since LDA underestimates the gap and HSE06 probably

overestimates it. Plots of the carrier concentration across the

transition are shown in Fig. 6 for both the LDA calculation

and the HSE06 (with PBE used for the metallic phase). If we

use the LDA þ U results to calculate the carrier concentra-tion changes (see Table III), we see that the relative jump

increases between the value of pure LDA and HSE06 as one

would expect since the band gap increases within the range

of the two. In this particular case, the fact that the relative

spacing of the valence band is wrong within LDA þ U playslittle role since the oxygen 2p derived states are energetically

very far from the conduction band relative to the thermal

energy even near the transition temperature. As the change

in the carrier concentration is the main figure of merit for the

practical applications of the metal-to-insulator transition in

NbO2, these results highlight the importance of correctly

determining the band gap of the low temperature BCT phase

of the material.

IV. OPTICAL PROPERTIES VIA ELLIPSOMETRY ANDTHEORY

A 36.8 nm-thick epitaxial film of NbO2 was grown on a

0.5 mm thick (111)-oriented (LaAlO3)0.3 (Sr2AlTaO6)0.7(LSAT) single crystal substrate using molecular beam epi-

taxy as discussed in previous work35 for use in ellipsometry

measurements. The resulting NbO2 films are oriented such

that they have (110) orientation out of plane and the c-axis in

plane. Due to the symmetry mismatch between the trigonal

LSAT (111) surface and the NbO2 (110) planes, the film

exhibits three symmetry-equivalent rotational domains. This

TABLE III. Summary of the change in magnitude of the carrier concentration calculated at the transition temperature (taken to be 1080 K). Acronyms for com-

putational details are explained in the text.

LDA þ U

LDA HSE06 U¼ 1.0 U¼ 2.0 U¼ 3.0 U¼ 4.0 U¼ 5.0

nBCT 3:71� 1020 1:02� 1018 2:48� 1020 1:18� 1020 5:08� 1019 1:68� 1019 4:63� 1018nrutile 4:61� 1021 4:77� 1021 4:35� 1021 3:85� 1021 3:24� 1021 2:57� 1021 1:38� 1021

log10nrutilenBCT

� �1.09 3.67 1.24 1.51 1.80 2.19 2.47

FIG. 6. Comparison of the carrier concentration calculated for a range of

temperatures crossing the transition temperature using both pure LDA and

HSE06 functionals.

213705-7 O’Hara et al. J. Appl. Phys. 116, 213705 (2014)

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.219.247.33 On: Tue, 20 Jan 2015 02:57:55

is confirmed by in situ reflection high energy electron diffrac-tion showing pseudo-six-fold azimuthal symmetry. In order

to confirm the phase purity of the NbO2 film, in situ XPSmeasurements were performed after growth using a VG

Scienta R3000 analyzer with monochromated Al Ka radia-tion (h�¼ 1486.6 eV). Valence band and Nb 3d core levelspectra were obtained using a pass energy of 100 eV and ana-

lyzer slit size of 0.4 mm, yielding a total resolution of

350 meV. The XPS spectra are shown in Figs. 7(a) and 7(b).

The valence band spectrum (Fig. 7(a)) shows two main fea-

tures: a �6 eV wide O 2p band and a �1 eV wide Nb 4dband with a height about 0.8 times that of the O 2p band. The

valence band spectrum is consistent with the calculated den-

sity of states for the low-temperature phase. The sharp Nb 4d

feature is centered at an energy �1.3 eV below the systemFermi level. The Nb 3d core level spectrum (Fig. 7(b)) can

be resolved into two components each consisting of a spin-

orbit pair with separation of 2.7 eV: a sharper feature at

206.0 eV and a broader feature at 207.0 eV, with an inte-

grated intensity ratio of 0.8 (I206/I207). These are convention-

ally assigned as originating from Nb4þ and Nb5þ oxidation

states, respectively, in the literature.57,58 However, we

believe that the multi-component nature of the Nb 3d core

level of NbO2 is due to final-state effects,59,60 and not to the

presence of Nb2O5. We cannot rule out, however, the exis-

tence of a thin Nb2O5 overlayer in the films measured with

ellipsometry as the uncapped NbO2 films have to be exposed

to air for several days prior to ellipsometric measurements.

Since the penetration depth of light in our NbO2 films ranges

from 10 nm in the UV to 100 nm in the infrared, the bulk of

the optical response in our spectra arises from the NbO2 film

and not from a (potential) thin Nb2O5 surface overlayer.

Also, we note that Nb2O5 has a band gap of 3.5 eV and a

broad absorption peak centered at 5 eV with a FWHM of

2 eV.61 No changes in the ellipsometry spectra were observed

if measurements were repeated several months apart.

Spectroscopic ellipsometry62,63 measures the complex

Fresnel ratio q ¼ rp=rs, where rp and rs are the complexreflectances for p- and s-polarized light. The Fresnel ratio is

usually expressed in terms of the ellipsometric angles W andD as q ¼ ðtan WÞeiD. For a smooth flat single-side polishedsubstrate without surface layers (such as roughness), one can

immediately determine the complex dielectric function of the

substrate.62 Corrections can be made for surface roughness.

This method was demonstrated for bulk MgAl2O4 spinel and

its optical constants have been published.64,65 The same

method was also used to determine the optical constants of

LSAT.65,66 For a flat NbO2 film (with a known thickness and

roughness, for example, determined by x-ray reflectance or

from the epitaxial growth conditions) grown on a substrate,

the optical constants of the NbO2 film can be determined

using standard ellipsometric data analysis techniques.62,63

For this work, we used two instruments: Data from 0.7

to 6.5 eV were acquired on a J.A. Woollam vertical variable-

angle-of-incidence ellipsometer (VASE) with a computer-

controlled Berek waveplate compensator. Data in the

infrared between 0.25 and 0.7 eV were acquired at room tem-

perature using a J.A. Woollam Fourier-transform infrared

VASE. Data from both instruments were merged and

analyzed simultaneously. This leads to a small discontinuity

in e1, which is not important. We used angles of incidencebetween 65� and 75�. Details are described elsewhere.67 Theresulting dielectric function for a 36.8 nm-thick film with a

surface roughness of 0.6 nm (determined by atomic force

microscopy) is shown in Fig. 7(c). We fitted the data at each

photon energy independently (point-by-point fit), which

increases the noise, but does not impose a chosen lineshape.

The Kramers-Kronig consistency of our data was verified

using oscillator fits, which yield similar results.62,63

Since NbO2 is not optically isotropic (i.e., not cubic),

we must address how the optical anisotropy affects our

ellipsometry results. At room temperature, NbO2 crystallizes

in a distorted rutile structure, which can be described as

body-centered tetragonal.35 As described above, the NbO2tetragonal [001] axis lies in the plane of the substrate and is

parallel to one of the three ½1�10� directions of LSAT.35 Inour ellipsometry experiment, even for large angles of inci-

dence, the refracted beam is nearly normal to the surface.

The electric field of the refracted light beam is therefore

nearly parallel to the surface. The optical axis of NbO2 lies

in the plane of the substrate, but the film consists of three

types of domains with different orientations relative to the

laboratory frame of the ellipsometer. Our experiment there-

fore measures the average of the ordinary and extraordinary

dielectric function of the NbO2 film. Peaks in our spectra

could originate in either the ordinary or extraordinary dielec-

tric function of NbO2 (or both). If peaks are shifted slightly

in the ordinary or extraordinary dielectric function, this will

lead to broadenings in our spectra.

In Fig. 7(c), we see that the imaginary part of the dielec-

tric function of NbO2 is small below 1 eV, consistent with

the behavior of an insulator. We observe peaks at 1.66, 3.95,

4.70, and 5.90 eV in e2 due to optical interband transitions.We can estimate the magnitude of the direct band gap of

NbO2 by plotting ðaEÞ2 versus photon energy E and extrapo-lating to zero, where a is the absorption coefficient. Theresults of this analysis are shown in Fig. 7(d). Using this pro-

cedure, it is always difficult to find a uniquely defined linear

region. Our best estimate (see Fig. 7(d)) yields a direct band

gap of 1.3 eV. The indirect band gap can be determined, in

theory, by plotting ðaEÞ1=2 versus photon energy, but this hasseveral practical problems. First, ellipsometry is able to mea-

sure large absorption coefficients (expected for direct band

gaps) much more accurately than the small absorption coeffi-

cients expected for indirect band gaps. This is even more

true for a very thin film (like our 36.8 nm-thick film of

NbO2). Second, indirect transitions are often assisted by sev-

eral phonons, which broaden indirect transitions, especially

at room temperature, which makes it even harder to deter-

mine a unique linear region.68 Despite these challenges, we

have used this method to analyze our data. As shown in Fig.

7(d), we find an indirect band gap of 0.7 eV, but this experi-

mental result should not be used to confirm the theoretical

result (mentioned earlier) that an indirect gap exists in NbO2.

For comparison with the ellipsometry measurements, we

compute the real and imaginary parts of the dielectric tensor

within the Kubo-Greenwood formalism. Within this approach,

the imaginary part of the dielectric tensor is given by

213705-8 O’Hara et al. J. Appl. Phys. 116, 213705 (2014)

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.219.247.33 On: Tue, 20 Jan 2015 02:57:55

FIG. 7. (a) XPS spectrum of the valence band for 36.8 nm of NbO2 grown on LSAT by MBE showing contributions from the O 2p and Nb 4d states. The strong

and sharp Nb 4d feature confirms the sample is mostly NbO2. (b) Niobium 3d core level spectrum consisting of two sets of spin-orbit-split pairs (206/209 eV

and 207/210 eV). The possible origins of the two components are discussed in the text. (c) Real and imaginary parts of the complex dielectric function for the

sample as a function of photon energy from 0.25 to 6.5 eV (bold). The real (dashed) and imaginary (dotted) parts of the dielectric function for Nb2O5 (Ref. 61)

are overlaid to rule out contributions to the structure from this material. (d) The direct band gap is estimated to be 1.3 eV by plotting ðaEÞ2 versus photonenergy E and extrapolating to zero. An indirect band gap might also exist near 0.7 eV, determined by extrapolating ðaEÞ1=2, but that is less accurate since ellips-ometry cannot measure the small absorption coefficient a expected for indirect transitions, especially for thin films. Data between 0.5 and 0.75 eV from FTIRellipsometry are noisy, but this does not change the conclusions. (e) The real and imaginary parts of the electronic dielectric tensor as a function of photon

energy computed with a Gaussian broadening of r¼ 0.55 eV and a scissor operator to open the indirect band gap to 0.95 eV. In addition to the average dielec-tric functions matched to the experimental orientation (solid lines), the two components e11 (dotted) and e33(dashed) are plotted to show the effect of crystallineanisotropy on the dielectric function.

213705-9 O’Hara et al. J. Appl. Phys. 116, 213705 (2014)

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.219.247.33 On: Tue, 20 Jan 2015 02:57:55

Im eab½ � ¼1

V

2pemx

� � X~k ;n1;n2

h~kn1jP_

aj~kn2ih~kn2jP_

bj~kn1i

� d E~kn2 � E~kn1 � �hx� �

; (3)

where a and b represent Cartesian directions, ~k is the latticewave-vector, the n1 are valence bands, the n2 are conductionbands, x is the photon frequency, and P̂ is the dipole matrixoperator. The dipole transition matrix elements can be calcu-

lated within the PAW formalism by utilizing the relationship

between the all-electron wave function and the pseudowave

function as well as the projector functions’ completeness

relation69,70

P̂n1n2 ¼ h ~Wn1 jP̂j ~Wn2i

þX

i;j

h ~Wn1 jpiiðh/ijP̂j/ji � h~/ijP̂j~/jiÞhpjj ~Wn2i;

(4)

where the ~W are the pseudo wave functions, the ~p are theprojectors, and / and ~/ are the AE and PAW partial waves.The calculation of these matrix elements has previously been

implemented within the VASP code.70 In order to obtain the

real part of the dielectric tensor from the imaginary part, we

employ the Kramers-Kronig relation

Re eab Ephð Þ½ � ¼ 1þ2

pPð1

0

E0phIm eab E0ph� ��

E02ph � E2phdE0ph; (5)

which we compute numerically using Maclaruin’s formula.71

The discretized integral is then computed by considering

only the odd (even) grid points of the imaginary part for an

even (odd) grid point of the real part. Such an approach

avoids the need to explicitly account for the singularity in

the denominator. For a tetragonal film, there are only two in-

dependent elements in the dielectric tensor, e11 and e33.Since the films are grown such that the tetragonal axis is in

the plane of the substrate and rutile-based films grown on

(111)-oriented cubic substrates exhibit three symmetry-

equivalent rotational domains, the measured values corre-

spond to the average of the two, e11þe332

. The k-point grid was

increased to 20� 20� 20 (1062 k-points within the irreduci-ble wedge of the Brillouin zone) in order to ensure conver-

gence of the dielectric functions.

As the LDA underestimates the indirect band gap (recall

we have calculated it to be 0.35 eV), to compare the experi-

mental and theoretical results directly, we apply a scissors

operator to the band structure of the insulating phase of

NbO2 (that is a rigid shift of the conduction band eigenval-

ues, which leaves the wave functions and hence matrix ele-

ments unaffected). We then adjust the band gap and apply

Gaussian broadening in an attempt to match the averaged

imaginary dielectric function, e2, to experiment. The locationof the first peak is best matched using an indirect gap of

0.925 eV (comparable to the experimental result for the indi-

rect band gap of 0.7 eV, see above), while the direct absorp-

tion edge in e2 is best matched with an indirect gap between0.975 and 1.00 eV. Overall, it appears that an indirect gap of

0.95 eV matches best. Recalling that within LDA, the lowest

direct gap is 0.07 eV above the indirect gap, this gives an

estimate for the direct gap to be 1.02 eV with an uncertainty

on the order of 0.1 eV. This value is then a slight underesti-

mate of the value obtained above for the direct gap from

ellipsometry accounting for the relative errors in both. The

theoretical result for this indirect gap with Gaussian broaden-

ing of r ¼ 0:55 eV is plotted in Fig. 7(e) along with the realpart of the dielectric function. We also note that the calcula-

tion gives a value for e1 (the zero energy limit of the elec-tronic part of e1) of 6.8, which is a slight underestimationcompared to experiment. Using the method described in Sec.

III E for calculating the carrier concentration in the insulat-

ing phase, we find that by applying the scissor operator with

an indirect gap of 0.95 eV results in ne ¼ 1:02� 1018cm�3at 1080 K, which is a change of 2.5 orders of magnitude at

the transition.

In Fig. 7(e), we also plot the real and imaginary parts of

e11 and e33 in addition to their average so that we can under-stand the role that the crystal’s tetragonal anisotropy plays.

Within the theoretical data, the imaginary part of the dielec-

tric function contains three primary features at 1.68, 4.21,

and 5.69 eV. There is also a feature at �2.68 eV, whichbarely appears due to the broadening used. In Fig. 7(e), these

features are labeled as A, C, D, and B, respectively. The pri-

mary low energy features in the ellipsometry data for the

imaginary part of the dielectric function aligns with the fea-

ture marked A in the calculated version. This feature primar-

ily originates from e11, but is also present to a lesser degreein e33. Using the band structure and density of states, it isclear that the only possible direct transitions that can be

attributed to this are those caused by excitations from the

split off dxy band to the remaining t2g bands. Interestingly, if

we consider the fact that the origin of the top of the valence

band is composed of a bonding state formed from the dxyorbitals, we would expect such a transition to be a forbidden

dipole transition using atomic selection rules. However, such

a feature occurs in the ellipsometry of VO2 as well,72 despite

the fact that the excitation has a long life time73 indicating

potential lack of a dipole transition. Given the location of the

oxygen 2p states relative to the niobium-niobium bonding

state, the feature labeled B on the theoretical plot would also

be comprised of excitations from the bonding state into the

remaining t2g orbitals. Interestingly, such a feature is not

readily observed in the experimental measurement and may

either be missed due to experimental broadening of the more

prominent lower and higher energy peaks or indicates a fur-

ther orbital-orientation-dependence not resolved in our cal-

culation. It is more clearly seen by considering e33 by itself.This question could be resolved with ellipsometry measure-

ments on different faces of a bulk NbO2 crystal, but not with

our films due to the three-fold symmetry of the LSAT (111)

surface.

In terms of transitions involving excitations only within

the d-band, the remaining features could be attributed to

excitations of the dxy bonding electrons into the antibonding

states (peak C) and into the niobium eg states (peak D).

However, we also recall that the top of the oxygen 2p states

sits approximately 1.58 eV below the bottom of the niobium

213705-10 O’Hara et al. J. Appl. Phys. 116, 213705 (2014)

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.219.247.33 On: Tue, 20 Jan 2015 02:57:55

dxy bonding state. This means that peaks C and D can both

be explained by excitations to the unoccupied t2g bands or

antibonding state from the oxygen 2p bands. Such transi-

tions, in general, are not dipole forbidden and therefore

would also help to explain why these excitations are stronger

both in experiment and theory. To clarify the source of these

peaks, we recalculated e11 and e33 using only the bandsbelonging to the oxygen 2p portion of the valence band and

then using only the bands belonging to the niobium dxy por-

tion of the valence band. The result of resolving the dielec-

tric function this way made it clear that peak C was only

comprised of the excitation of electrons from the dxy bonding

state into the antibonding state and that the for peak D, both

the e11 and e33 contributions come from excitations of the ox-ygen 2p electrons into the unoccupied t2g bands (there is a

very minor contribution to e11 from dxy to eg excitations).Any excitations from the oxygen states into the niobium egstates would be higher energy excitations than are plotted in

the figures. Finally, we note that while our main peaks A, C,

and D align roughly with those of the measurements, the cal-

culation does not contain a peak near the 4.7 eV experimen-

tal peak. The origin may be due to a physical effect

neglected by standard LDA and requiring a more advanced

theory.

V. CONCLUSIONS

We report on the electronic structure calculations for the

high and low temperature phases of NbO2 using density

functional theory. While the local density approximation

yields a value of the band gap well below reported values,

calculations using the HSE06 hybrid potential overestimate

the correction. Furthermore, pure LDA appears to be more

consistent with experiment than HSE06 in terms of the

change in magnitude of the carrier concentration upon heat-

ing across the transition temperature. Calculations with a

Hubbard type U correction show that small values of U in

the range of 0 eV to 2 eV may slightly improve the lattice pa-

rameters of the insulating phase as well as increase the band

gap, however, at the expense of correctly describing the oxy-

gen 2p–niobium dxy gap. Full Brillouin zone Fermi surface

calculations show evidence of nesting consistent with wave-

vectors related to dimerization. Spectroscopic ellipsometry

performed on a 36.8 nm-thick epitaxial film of NbO2 shows

that the lowest direct gap occurs at 1.3 eV. An indirect gap

may exist below 1 eV. A theoretical calculation of the dielec-

tric function including matrix element effects shows that

absorption onset and peak placement are consistent with an

indirect band gap of 0.95 eV, which would predict a change

in carrier concentration of 2.5 orders of magnitude at the

metal-insulator transition.

ACKNOWLEDGMENTS

This work was supported by the SRC Contract 2012-VJ-

2299, and Texas Advanced Computing Center (TACC). The

work at NMSU was supported by the National Science

Foundation (DMR-1104934). The FTIR ellipsometry

measurements were performed at the Center for Integrated

Nanotechnologies, an Office of Science User Facility

operated for the U.S. Department of Energy (DOE) Office of

Science by Sandia National Laboratory (Contract No. DE-

AC04-94AL85000). We thank Ilya Karpov and Chungwei

Lin for helpful discussions.

1W. Fogle and J. H. Perlstein, Phys. Rev. B 6, 1402 (1972).2M. Greenblatt, Chem. Rev. 88, 31 (1988).3C. Sellier, F. Boucher, and E. Janod, Solid State Sci. 5, 591 (2003).4P. M. Marley, A. A. Stabile, C. P. Kwan, S. Singh, P. Zhang, G.

Sambandamurthy, and S. Banerjee, Adv. Funct. Mater. 23, 153 (2013).5J. Torrence, P. Lacorre, and A. Nazzal, Phys. Rev. B 45, 8209 (1992).6M. L. Medarde, J. Phys.: Condens. Matter 9, 1679 (1997).7J. A. Alonso, M. J. Mart�ınez-Lope, J. L. Garc�ıa-Mu~noz, and M. T.Fern�andez-D�ıaz, Phys. Rev. B 61, 1756 (2000).

8A. Cavalleri, T. Dekorsy, H. Chong, J. Kieffer, and R. Schoenlein, Phys.

Rev. B 70, 161102 (2004).9S. Biermann, A. Poteryaev, A. Lichtenstein, and A. Georges, Phys. Rev.

Lett. 94, 26404 (2005).10M. M. Qazilbash, M. Brehm, B.-G. Chae, P.-C. Ho, G. O. Andreev, B.-J.

Kim, S. J. Yun, A. V. Balatsky, M. B. Maple, F. Keilmann, H.-T. Kim,

and D. N. Basov, Science 318, 1750 (2007).11E. Strelcov, Y. Lilach, and A. Kolmakov, Nano Lett. 9, 2322 (2009).12A. C. Jones, S. Berweger, J. Wei, D. Cobden, and M. B. Raschke, Nano

Lett. 10, 1574 (2010).13R. Janninck and D. Whitmore, J. Phys. Chem. Solids 27, 1183 (1966).14G. Belanger, J. Destry, G. Perluzzo, and P. M. Raccah, Can. J. Phys. 52,

2272 (1974); available at http://www.nrcresearchpress.com/doi/abs/

10.1139/p74-297#.VHTVM4vF9Og.15S. H. Shin, T. Halpern, and P. M. Raccah, J. Appl. Phys. 48, 3150 (1977).16H. R. Philipp and L. M. Levinson, J. Appl. Phys. 50, 4814 (1979).17J. C. Lee and W. W. Durand, J. Appl. Phys. 56, 3350 (1984).18S. Kim, J. Park, J. Woo, C. Cho, W. Lee, J. Shin, G. Choi, S. Park, D. Lee,

B. H. Lee, and H. Hwang, Microelectron. Eng. 107, 33 (2013).19E. Cha, J. Woo, D. Lee, S. Lee, J. Song, Y. Koo, J. Lee, C. G. Park, M. Y.

Yang, K. Kamiya, K. Shiraishi, B. Magyari-Kope, Y. Nishi, and H.

Hwang, in 2013 IEEE International Electron Devices Meeting (IEEE,2013), pp. 10.5.1–10.5.4.

20F. Morin, Phys. Rev. Lett. 3, 34 (1959).21R. Pynn and J. D. Axe, J. Phys. C: Solid State Phys. 9, L199 (1976).22K. Seta and K. Naito, J. Chem. Thermodyn. 14, 921 (1982).23A. K. Cheetham and C. N. R. Rao, Acta Crystallogr., Sect. B: Struct.

Crystallogr. Cryst. Chem. 32, 1579 (1976).24R. Pynn, J. Axe, and R. Thomas, Phys. Rev. B 13, 2965 (1976).25A. A. Bolzan, C. Fong, B. J. Kennedy, and C. J. Howard, J. Solid State

Chem. 113, 9 (1994).26S. M. Shapiro, J. D. Axe, G. Shirane, and P. M. Raccah, Solid State

Commun. 15, 377 (1974).27R. Pynn, J. Axe, and P. Raccah, Phys. Rev. B 17, 2196 (1978).28F. Gervais and W. Kress, Phys. Rev. B 31, 4809 (1985).29C. N. R. Rao, G. R. Rao, and G. V. S. Rao, J. Solid State Chem. 6, 340 (1973).30K. T. Jacob, C. Shekhar, M. Vinay, and Y. Waseda, J. Chem. Eng. Data

55, 4854 (2010).31A. O’Hara and A. A. Demkov, “The nature of the metal-insulator transi-

tion in NbO2” (unpbulished).32D. Adler, Rev. Mod. Phys. 40, 714 (1968).33M. Yousuf, J. Appl. Phys. 53, 8647 (1982).34Y. Sakai, N. Tsuda, and T. Sakata, J. Phys. Soc. Jpn. 54, 1514 (1985).35A. B. Posadas, A. O’Hara, S. Rangan, R. A. Bartynski, and A. A. Demkov,

Appl. Phys. Lett. 104, 092901 (2014).36F. J. Wong, N. Hong, and S. Ramanathan, Phys. Rev. B 90, 115135

(2014).37G. Kresse and J. Furthm€uller, Phys. Rev. B 54, 11169 (1996).38J. Perdew and A. Zunger, Phys. Rev. B 23, 5048 (1981).39P. Bl€ochl, Phys. Rev. B 50, 17953 (1994).40G. Kresse and D. Joubert, Phys. Rev. B 59, 11 (1999).41H. Monkhorst and J. Pack, Phys. Rev. B 13, 5188 (1976).42M. Methfessel and A. Paxton, Phys. Rev. B 40, 3616 (1989).43P. Bl€ochl, O. Jepsen, and O. Andersen, Phys. Rev. B 49, 16223 (1994).44M. Posternak, A. Freeman, and D. Ellis, Phys. Rev. B 19, 6555 (1979).45T. A. Sasaki and T. Soga, Physica BþC 111, 304 (1981).46V. Eyert, Europhys. Lett. 58, 851 (2002).47S. Kagoshima, H. Nagasawa, and T. Sambongi, One-Dimensional

Conductors (Springer-Verlag, Berlin, 1982), pp. 20–21.

213705-11 O’Hara et al. J. Appl. Phys. 116, 213705 (2014)

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.219.247.33 On: Tue, 20 Jan 2015 02:57:55

http://dx.doi.org/10.1103/PhysRevB.6.1402http://dx.doi.org/10.1021/cr00083a002http://dx.doi.org/10.1016/S1293-2558(03)00045-1http://dx.doi.org/10.1002/adfm.201201513http://dx.doi.org/10.1103/PhysRevB.45.8209http://dx.doi.org/10.1088/0953-8984/9/8/003http://dx.doi.org/10.1103/PhysRevB.61.1756http://dx.doi.org/10.1103/PhysRevB.70.161102http://dx.doi.org/10.1103/PhysRevB.70.161102http://dx.doi.org/10.1103/PhysRevLett.94.026404http://dx.doi.org/10.1103/PhysRevLett.94.026404http://dx.doi.org/10.1126/science.1150124http://dx.doi.org/10.1021/nl900676nhttp://dx.doi.org/10.1021/nl903765hhttp://dx.doi.org/10.1021/nl903765hhttp://dx.doi.org/10.1016/0022-3697(66)90094-1http://www.nrcresearchpress.com/doi/abs/10.1139/p74-297#.VHTVM4vF9Oghttp://www.nrcresearchpress.com/doi/abs/10.1139/p74-297#.VHTVM4vF9Oghttp://dx.doi.org/10.1063/1.324047http://dx.doi.org/10.1063/1.326544http://dx.doi.org/10.1063/1.333863http://dx.doi.org/10.1016/j.mee.2013.02.084http://dx.doi.org/10.1103/PhysRevLett.3.34http://dx.doi.org/10.1088/0022-3719/9/8/003http://dx.doi.org/10.1016/0021-9614(82)90002-7http://dx.doi.org/10.1107/S0567740876005876http://dx.doi.org/10.1107/S0567740876005876http://dx.doi.org/10.1103/PhysRevB.13.2965http://dx.doi.org/10.1006/jssc.1994.1334http://dx.doi.org/10.1006/jssc.1994.1334http://dx.doi.org/10.1016/0038-1098(74)90780-7http://dx.doi.org/10.1016/0038-1098(74)90780-7http://dx.doi.org/10.1103/PhysRevB.17.2196http://dx.doi.org/10.1103/PhysRevB.31.4809http://dx.doi.org/10.1016/0022-4596(73)90219-3http://dx.doi.org/10.1021/je1004609http://dx.doi.org/10.1103/RevModPhys.40.714http://dx.doi.org/10.1063/1.330461http://dx.doi.org/10.1143/JPSJ.54.1514http://dx.doi.org/10.1063/1.4867085http://dx.doi.org/10.1103/PhysRevB.90.115135http://dx.doi.org/10.1103/PhysRevB.54.11169http://dx.doi.org/10.1103/PhysRevB.23.5048http://dx.doi.org/10.1103/PhysRevB.50.17953http://dx.doi.org/10.1103/PhysRevB.59.1758http://dx.doi.org/10.1103/PhysRevB.13.5188http://dx.doi.org/10.1103/PhysRevB.40.3616http://dx.doi.org/10.1103/PhysRevB.49.16223http://dx.doi.org/10.1103/PhysRevB.19.6555http://dx.doi.org/10.1016/0378-4363(81)90108-Xhttp://dx.doi.org/10.1016/0378-4363(81)90108-Xhttp://dx.doi.org/10.1209/epl/i2002-00452-6

48N. Beatham and A. F. Orchard, J. Electron Spectrosc. Relat. Phenom. 16,77 (1979).

49N. Jiang and J. Spence, Phys. Rev. B 70, 245117 (2004).50J. Goodenough, Phys. Rev. 117, 1442 (1960).51S. Dudarev and G. Botton, Phys. Rev. B 57, 1505 (1998).52G.-H. Liu, X.-Y. Deng, and R. Wen, J. Mater. Sci. 45, 3270 (2010).53J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 118, 8207

(2003).54J. Heyd, G. E. Scuseria, and M. Ernzerhof, J. Chem. Phys. 124, 219906

(2006).55C. J. Cramer and D. G. Truhlar, Phys. Chem. Chem. Phys. 11, 10757

(2009).56J. Perdew, K. Burke, and M. Ernzerhof, Phys. Rev. Lett. 77, 3865 (1996).57M. Kuznetsov, A. Razinkin, and E. Shalaeva, J. Struct. Chem. 50, 536

(2009).58M. K. Bahl, J. Phys. Chem. Solids 36, 485 (1975).59D. Morris, Y. Dou, J. Rebane, C. Mitchell, R. Egdell, D. Law, A.

Vittadini, and M. Casarin, Phys. Rev. B 61, 13445 (2000).60N. Beatham, P. A. Cox, R. G. Egdell, and A. F. Orchard, Chem. Phys.

Lett. 69, 479 (1980).61M. Ser�enyi, T. Lohner, P. Petrik, Z. Zolnai, Z. E. Horv�ath, and N. Q.

Kh�anh, Thin Solid Films 516, 8096 (2008).

62H. G. Tompkins and W. A. McGahan, Spectroscopic Ellipsometry andReflectometry: A User’s Guide (Wiley, New York, 1999).

63H. Fujiwara, Spectroscopic Ellipsometry (Wiley, Chichester, UK, 2007).64C. J. Zollner, T. I. Willett-Gies, S. Zollner, and S. Choi, “Infrared to

vacuum-ultraviolet ellipsometry studies of spinel (MgAl2O4),” Thin Solid

Films (published online).65S. Zollner, “The Ellipsometry Group at New Mexico State University” see

http://ellipsometry.nmsu.edu/.66T. N. Nunley, T. I. Willett-Gies, and S. Zollner, in Rio Grande

Symposium on Advanced Materials, Albuquerque, NM, 2013.67K. J. Kormondy, A. B. Posadas, A. Slepko, A. Dhamdhere, D. J. Smith, K.

N. Mitchell, T. I. Willett-Gies, S. Zollner, L. G. Marshall, J. Zhou, and A.

A. Demkov, J. Appl. Phys. 115, 243708 (2014).68P. Yu and M. Cardona, Fundamentals of Semiconductors (Springer,

Heidelberg, 2010).69H. Kageshima and K. Shiraishi, Phys. Rev. B 56, 14985 (1997).70B. Adolph, J. Furthm€uller, and F. Bechstedt, Phys. Rev. B 63, 125108 (2001).71K. Ohta and H. Ishida, Appl. Spectrosc. 42, 952 (1988).72M. M. Qazilbash, A. A. Schafgans, K. S. Burch, S. J. Yun, B. G. Chae, B.

J. Kim, H. T. Kim, and D. N. Basov, Phys. Rev. B 77, 115121 (2008).73C. Miller, M. Triplett, J. Lammatao, J. Suh, D. Fu, J. Wu, and D. Yu,

Phys. Rev. B 85, 085111 (2012).

213705-12 O’Hara et al. J. Appl. Phys. 116, 213705 (2014)

[This article is copyrighted as indicated in the article. Reuse of AIP content is subject to the terms at: http://scitation.aip.org/termsconditions. Downloaded to ] IP:

129.219.247.33 On: Tue, 20 Jan 2015 02:57:55

http://dx.doi.org/10.1016/0368-2048(79)85006-9http://dx.doi.org/10.1103/PhysRevB.70.245117http://dx.doi.org/10.1103/PhysRev.117.1442http://dx.doi.org/10.1103/PhysRevB.57.1505http://dx.doi.org/10.1007/s10853-010-4338-2http://dx.doi.org/10.1063/1.1564060http://dx.doi.org/10.1063/1.2204597http://dx.doi.org/10.1039/b907148bhttp://dx.doi.org/10.1103/PhysRevLett.77.3865http://dx.doi.org/10.1007/s10947-009-0079-yhttp://dx.doi.org/10.1016/0022-3697(75)90132-8http://dx.doi.org/10.1103/PhysRevB.61.13445http://dx.doi.org/10.1016/0009-2614(80)85108-6http://dx.doi.org/10.1016/0009-2614(80)85108-6http://dx.doi.org/10.1016/j.tsf.2008.04.070http://dx.doi.org/10.1016/j.tsf.2013.11.141http://dx.doi.org/10.1016/j.tsf.2013.11.141http://ellipsometry.nmsu.edu/http://dx.doi.org/10.1063/1.4885048http://dx.doi.org/10.1103/PhysRevB.56.14985http://dx.doi.org/10.1103/PhysRevB.63.125108http://dx.doi.org/10.1366/0003702884430380http://dx.doi.org/10.1103/PhysRevB.77.115121http://dx.doi.org/10.1103/PhysRevB.85.085111